Our lab typically studies high-level visual cortex, but in the past year we published work on a property that seems to be a guiding principle of neural representations in the first station of visual cortical processing, the primary visual cortex or V1. We found that deep neural networks that have straightened representations for natural movies were the best at explaining neural data from V1 (and V2). To produce such straightened representations, networks needed to be trained with noisy images. Thus, robustness to noise, a biologically plausible improvement to deep neural networks, may lead to better models of early visual cortex.

We just initiated high-density neural recordings in the primate (common marmoset) inferior temporal cortex using neuropixels probes. We are able to record from 100-200 neurons in a session and are sampling many visual areas across the inferior temporal lobe of the small, flat marmoset brain which is amenable to physiological mapping. We are looking forward to learning about the many subregions of the inferior temporal lobe from face- and object-selective IT cortex to the scene-selective parahippocampal cortex and into the hippocampus which is involved in visually guided spatial navigation in primates. We plan to test our hypothesis about prediction versus discrimination objectives in the visual system in these neural structures.

The Grossman Center brings together a critical mass of primate researchers such that we can share methods (neuropixels probes) as well as scientific ideas (high-level vision and cognition; population neural dynamics). For example, I was recently talking with Dr. Churchland about the unique flexibility of population decoders in very high-dimensional spaces in his motor cortex studies, and I realized we had likely observed similar phenomena in the visual system. Interactions at these different levels are crucial for driving primate research forward. The reward is that we can study the neural bases of human-like behavior in homologous brain areas in primates, such as face discrimination in face-selective cortex or visually guided spatial reasoning in the medial temporal lobe.

Recently, I initiated a collaboration with Liam Paninski’s group to apply their neural data analysis tools to our neuropixels recordings from marmosets. Indeed, I found that they had already implemented many unique statistical analyses that would provide us with unprecedented insights from our datasets. These sorts of mutually beneficial research directions are continuously happening here, and the Grossman Center’s environment allows us to tap their potential.

As I conclude my tenure as a Grossman-Kavli Scholar, I would like to say it has truly been an honor and a pleasure to be affiliated with the Grossman Center.